OpenAI has released plugins for ChatGPT to enable it access third-party knowledge sources and databases, with the internet inclusive.
OpenAI asserts it will emphasize a few developers and subscribers to its premium ChatGPT Plus plan before wheeling out larger-scale and API access in alpha to ChatGPT users and developers on the waitlist.
The most fascinating plugin, by far, is OpenAI’s first-party web-browsing plugin, which enables ChatGPT to draw data from the web to reply to the endless questions presented to it.
At inception, ChatGPT’s knowledge base was restricted to dates, events, and people before September 2021 The plugin uses the Bing search API to obtain content from the web and displays any websites it visited while crafting an answer, citing its sources in ChatGPT’s replies.
According to OpenAI’s research, a chatbot with web access is a risky proposition. An experimental platform designed in 2021 by the AI startup, called WebGPT, occasionally quoted from questionable sources and was incentivized to choose and take data from sites it presumed users would find persuasive even if those sources weren’t factually the best.
Meta’s since-disbanded BlenderBot 3.0 had access to the web, too, and immediately went off the beams, venturing into conspiracy theories and inappropriate contents when urged with certain text.
The live web is less curated and, implicitly, less filtered than the static training data set. Search engines such as Google and Bing use their own security mechanisms to curtail the chances of false content appearing at the top of results, but still allow those results to be played.
Nor are they no doubt representative of the web as a whole. According to The New Yorker, Google’s algorithms favor sites that use modern web technologies like encryption, mobile support, and schema markup. As a result, many sites that offer quality content end up cluttered.
This provides search engines a lot of control over the data that could be used to notify the replies of web-connected language models. Google has been discovered to compute its own services in Search, such as responding to a travel inquiry with data from Google Places rather than a richer, more social source such as TripAdvisor.
Simultaneously, the algorithmic approach to search invites bad actors. According to The New Yorker, Pinterest used a bug in Google’s image search algorithm in 2020 to reveal more of its content in Google Image searches.
OpenAI accepts that web-enabled ChatGPT can perform all sorts of unwanted behavior. For example, sending fake or spam emails, circumventing security restrictions, and generally “improving the skills of malicious individuals who fraud, mislead, or abuse others.
In addition to the web plugin, OpenAI has released a code interpreter for his ChatGPT. It provides your chatbot with a Python interpreter that runs in a sandbox environment with firewalls and disk space.
Various early contributors, including Expedia, FiscalNote, Instacart, Kayak, Klarna, Milo, OpenTable, Shopify, Slack, Speak, Wolfram, and Zapier, have created plugins for ChatGPT to join OpenAI.
They are mostly self-explanatory. For example, the OpenTable plugin allows a chatbot to see which restaurants are available for reservations.
On the other hand, the Instacart plugin allows you to place orders from local businesses on ChatGPT. The most scalable, Zapier connects to apps like Google Sheets, Trello, and Gmail to trigger a variety of productivity tasks.
To facilitate the creation of new plugins, OpenAI provided an open-source “fetch” plugin. This allows ChatGPT to access snippets of documents from data sources such as files, notes, emails, and public documents by asking questions in natural language.
The plugin is a strange addition to the ChatGPT development timeline. Limited to information from training data, ChatGPT suddenly becomes more powerful with plugins, and possibly less legal risk.
Some experts have accused OpenAI of profiting from the unauthorized work his ChatGPT trained for. ChatGPT’s dataset includes a variety of public websites. But plugins could solve this problem by giving companies full control over their data.
Don’t miss important articles during the week. Subscribe to techbuild.africa weekly digest for updates